Abstract: In order to meet the requirements of preprocessing of Chinese cheque image recognition, this paper studied several binarization algorithms. After analyzing the histograms of gray-level images of 2000 cheques, we found a clue,maximum gradient, in histogram of gray-level cheque image, which can be used for segmenting image and extracting given object parallel-line of amount field. With the clue, we developed a new binarization algorithm which makes the parallel-line of amount field more obvious and easier to be located in binarization image. Comparing with several other commonly used binarization algorithms,the algorithm proposed by the paper has been proven to be more feasible and advanced in the simulating tests.

Key words: Chinese checaue image recognition, Binarization, Histogram, Gradient